DPDK patches and discussions
 help / color / mirror / Atom feed
From: Ariel Otilibili <ariel.otilibili@6wind.com>
To: Maryam Tahhan <mtahhan@redhat.com>
Cc: "dev@dpdk.org" <dev@dpdk.org>,
	"stable@dpdk.org" <stable@dpdk.org>,
	Thomas Monjalon <thomas@monjalon.net>,
	 David Marchand <david.marchand@redhat.com>,
	Stephen Hemminger <stephen@networkplumber.org>,
	 Ciara Loftus <ciara.loftus@intel.com>
Subject: Re: [PATCH v7 2/2] net/af_xdp: Refactor af_xdp_tx_zc
Date: Thu, 6 Feb 2025 19:06:04 +0100	[thread overview]
Message-ID: <CAF1zDgZ18Ov5ADzcFLgZoQB5RMVd8w=8-KpHQV6bGSe_QX44bg@mail.gmail.com> (raw)
In-Reply-To: <CAFdtZiu9t6tzw=cZgN+B22DiC7eYbMx7hxmBzVbx9Mfz3XwMVQ@mail.gmail.com>

[-- Attachment #1: Type: text/plain, Size: 6900 bytes --]

Hi Maryam,

On Thu, Feb 6, 2025 at 2:09 AM Maryam Tahhan <mtahhan@redhat.com> wrote:

>
>
> On Sat 1 Feb 2025, 10:03 Ariel Otilibili, <ariel.otilibili@6wind.com>
> wrote:
>
>> Both legs of the loop share the same logic: the common parts are about
>> reserving and filling both address and length into the description.
>>
>> This is moved into reserve_and_fill().
>>
>> Bugzilla ID: 1440
>> Suggested-by: Maryam Tahhan <mtahhan@redhat.com>
>> Signed-off-by: Ariel Otilibili <ariel.otilibili@6wind.com>
>> ---
>>  drivers/net/af_xdp/rte_eth_af_xdp.c | 62 ++++++++++++++++-------------
>>  1 file changed, 34 insertions(+), 28 deletions(-)
>>
>> diff --git a/drivers/net/af_xdp/rte_eth_af_xdp.c
>> b/drivers/net/af_xdp/rte_eth_af_xdp.c
>> index 092bcb73aa0a..840a12dbf508 100644
>> --- a/drivers/net/af_xdp/rte_eth_af_xdp.c
>> +++ b/drivers/net/af_xdp/rte_eth_af_xdp.c
>> @@ -536,6 +536,31 @@ kick_tx(struct pkt_tx_queue *txq, struct
>> xsk_ring_cons *cq)
>>                 }
>>  }
>>
>> +static inline struct xdp_desc *
>> +reserve_and_fill(struct pkt_tx_queue *txq, struct rte_mbuf *mbuf,
>> +                struct xsk_umem_info *umem)
>> +{
>> +       struct xdp_desc *desc = NULL;
>> +       uint32_t *idx_tx = NULL;
>> +       uint64_t addr, offset;
>> +
>> +       if (!xsk_ring_prod__reserve(&txq->tx, 1, idx_tx))
>> +               goto out;
>> +
>> +       desc = xsk_ring_prod__tx_desc(&txq->tx, *idx_tx);
>> +       desc->len = mbuf->pkt_len;
>> +
>> +       addr = (uint64_t)mbuf - (uint64_t)umem->buffer
>> +               - umem->mb_pool->header_size;
>> +       offset = rte_pktmbuf_mtod(mbuf, uint64_t) - (uint64_t)mbuf
>> +               + umem->mb_pool->header_size;
>> +       offset = offset << XSK_UNALIGNED_BUF_OFFSET_SHIFT;
>> +       desc->addr = addr | offset;
>> +
>> +out:
>> +       return desc;
>> +}
>> +
>>  #if defined(XDP_UMEM_UNALIGNED_CHUNK_FLAG)
>>  static uint16_t
>>  af_xdp_tx_zc(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
>> @@ -545,10 +570,8 @@ af_xdp_tx_zc(void *queue, struct rte_mbuf **bufs,
>> uint16_t nb_pkts)
>>         struct rte_mbuf *mbuf;
>>         unsigned long tx_bytes = 0;
>>         int i;
>> -       uint32_t idx_tx;
>>         uint16_t count = 0;
>>         struct xdp_desc *desc;
>> -       uint64_t addr, offset;
>>         struct xsk_ring_cons *cq = &txq->pair->cq;
>>         uint32_t free_thresh = cq->size >> 1;
>>
>> @@ -559,21 +582,12 @@ af_xdp_tx_zc(void *queue, struct rte_mbuf **bufs,
>> uint16_t nb_pkts)
>>                 mbuf = bufs[i];
>>
>>                 if (mbuf->pool == umem->mb_pool) {
>> -                       if (!xsk_ring_prod__reserve(&txq->tx, 1,
>> &idx_tx)) {
>> +                       if (!(desc = reserve_and_fill(txq, mbuf, umem))) {
>>                                 kick_tx(txq, cq);
>> -                               if (!xsk_ring_prod__reserve(&txq->tx, 1,
>> -                                                           &idx_tx))
>> +                               if (!(desc = reserve_and_fill(txq, mbuf,
>> umem)))
>>                                         goto out;
>>                         }
>> -                       desc = xsk_ring_prod__tx_desc(&txq->tx, idx_tx);
>> -                       desc->len = mbuf->pkt_len;
>> -                       addr = (uint64_t)mbuf - (uint64_t)umem->buffer -
>> -                                       umem->mb_pool->header_size;
>> -                       offset = rte_pktmbuf_mtod(mbuf, uint64_t) -
>> -                                       (uint64_t)mbuf +
>> -                                       umem->mb_pool->header_size;
>> -                       offset = offset << XSK_UNALIGNED_BUF_OFFSET_SHIFT;
>> -                       desc->addr = addr | offset;
>> +
>>                         tx_bytes += desc->len;
>>                         count++;
>>                 } else {
>> @@ -584,26 +598,18 @@ af_xdp_tx_zc(void *queue, struct rte_mbuf **bufs,
>> uint16_t nb_pkts)
>>                         if (local_mbuf == NULL)
>>                                 goto out;
>>
>> -                       if (!xsk_ring_prod__reserve(&txq->tx, 1,
>> &idx_tx)) {
>> +                       if (!(desc = reserve_and_fill(txq, local_mbuf,
>> umem))) {
>>                                 rte_pktmbuf_free(local_mbuf);
>>                                 goto out;
>>                         }
>>
>> -                       desc = xsk_ring_prod__tx_desc(&txq->tx, idx_tx);
>> -                       desc->len = mbuf->pkt_len;
>> -
>>
>
> So I think i spotted one issue, you might need override desc->len after
> the call to the reserve_and_fill function so as it's not taken supposed to
> be from local_mbuf here.
>

You are right; I will do that.

>
> -                       addr = (uint64_t)local_mbuf -
>> (uint64_t)umem->buffer -
>> -                                       umem->mb_pool->header_size;
>> -                       offset = rte_pktmbuf_mtod(local_mbuf, uint64_t) -
>> -                                       (uint64_t)local_mbuf +
>> -                                       umem->mb_pool->header_size;
>> -                       pkt = xsk_umem__get_data(umem->buffer, addr +
>> offset);
>> -                       offset = offset << XSK_UNALIGNED_BUF_OFFSET_SHIFT;
>> -                       desc->addr = addr | offset;
>> +                       pkt = xsk_umem__get_data(umem->buffer,
>> +                                                (desc->addr & ~0xFFF)
>> +                                                + (desc->addr & 0xFFF));
>>
>
> Would prefer to move this pkt assignment to reserve_and_fill()
> What if we passed a void **ppkt to reserve_and_fill()  then kept the
> original logic just wrapped in a NULL check?
>

It would look neater, indeed.

>
>  if (ppkt) {
>       *ppkt = xsk_umem__get_data(umem->buffer, addr + offset);
>     }
>
>
> Before the offset and desc->addr
>     offset = offset << XSK_UNALIGNED_BUF_OFFSET_SHIFT;
>     desc->addr = addr | offset;
>
> WDYT?
>

I will; thanks for the proposal.

For me to sum up which changes are needed:

* change desc->len to local_buf
* move pkt initialization into reserve_and_fill
*  change idx_tx from uint32_t* to uint32_t (
https://inbox.dpdk.org/dev/CAF1zDgZgJZgWOeG=RAjicmGVt24PTtB9B0ukWH_RdE3ooeYPzA@mail.gmail.com/
)

Thanks again,
Ariel

>
>                         rte_memcpy(pkt, rte_pktmbuf_mtod(mbuf, void *),
>> -                                       desc->len);
>> -                       tx_bytes += desc->len;
>> +                                  desc->len);
>>                         rte_pktmbuf_free(mbuf);
>> +                       tx_bytes += desc->len;
>>                         count++;
>>                 }
>>         }
>> --
>> 2.30.2
>>
>>

[-- Attachment #2: Type: text/html, Size: 10351 bytes --]

  parent reply	other threads:[~2025-02-06 18:06 UTC|newest]

Thread overview: 45+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2025-01-16 19:56 [PATCH 0/2] Fix use after free, and refactor af_xdp_tx_zc() Ariel Otilibili
2025-01-16 19:56 ` [PATCH 1/2] net/af_xdp: fix use after free in af_xdp_tx_zc() Ariel Otilibili
2025-01-30 18:24   ` Stephen Hemminger
2025-01-30 22:22     ` Ariel Otilibili
2025-01-16 19:56 ` [PATCH 2/2] net/af_xdp: Refactor af_xdp_tx_zc() Ariel Otilibili
2025-01-16 21:47   ` Stephen Hemminger
2025-01-16 22:20     ` Ariel Otilibili
2025-01-16 22:26       ` Stephen Hemminger
2025-01-16 22:36         ` Ariel Otilibili
2025-01-16 22:51 ` [PATCH v2 0/2] Fix use after free, and refactor af_xdp_tx_zc() Ariel Otilibili
2025-01-16 22:51   ` [PATCH v2 1/2] net/af_xdp: fix use after free in af_xdp_tx_zc() Ariel Otilibili
2025-01-20 14:54     ` Maryam Tahhan
2025-01-21  8:54       ` Ariel Otilibili
2025-01-16 22:51   ` [PATCH v2 2/2] net/af_xdp: Refactor af_xdp_tx_zc() Ariel Otilibili
2025-01-20 15:28     ` Maryam Tahhan
2025-01-21  8:57       ` Ariel Otilibili
2025-01-28 23:11 ` [PATCH v3 0/2] Fix use after free, and refactor af_xdp_tx_zc() Ariel Otilibili
2025-01-28 23:11   ` [PATCH v3 1/2] net/af_xdp: fix use after free in af_xdp_tx_zc() Ariel Otilibili
2025-01-28 23:11   ` [PATCH v3 2/2] net/af_xdp: Refactor af_xdp_tx_zc() Ariel Otilibili
2025-01-29 17:58     ` Maryam Tahhan
2025-01-30 23:06       ` Ariel Otilibili
2025-01-30 22:18 ` [PATCH v4 0/2] Fix use after free, and refactor af_xdp_tx_zc Ariel Otilibili
2025-01-30 22:18   ` [PATCH v4 1/2] net/af_xdp: Fix use after free in af_xdp_tx_zc Ariel Otilibili
2025-01-30 22:18   ` [PATCH v4 2/2] net/af_xdp: Refactor af_xdp_tx_zc Ariel Otilibili
2025-01-30 23:55     ` Stephen Hemminger
2025-01-31 18:36       ` Ariel Otilibili
2025-01-31 18:34 ` [PATCH v5 0/2] Fix use after free, and refactor af_xdp_tx_zc Ariel Otilibili
2025-01-31 18:34   ` [PATCH v5 1/2] net/af_xdp: Fix use after free in af_xdp_tx_zc Ariel Otilibili
2025-01-31 18:34   ` [PATCH v5 2/2] net/af_xdp: Refactor af_xdp_tx_zc Ariel Otilibili
2025-01-31 23:10 ` [PATCH v6 0/2] Fix use after free, and refactor af_xdp_tx_zc Ariel Otilibili
2025-01-31 23:10 ` [PATCH v6 1/2] net/af_xdp: Fix use after free in af_xdp_tx_zc Ariel Otilibili
2025-01-31 23:10 ` [PATCH v6 2/2] net/af_xdp: Refactor af_xdp_tx_zc Ariel Otilibili
2025-02-01 10:02 ` [PATCH v7 0/2] Fix use after free, and refactor af_xdp_tx_zc Ariel Otilibili
2025-02-01 10:02   ` [PATCH v7 1/2] net/af_xdp: Fix use after free in af_xdp_tx_zc Ariel Otilibili
2025-02-01 10:03   ` [PATCH v7 2/2] net/af_xdp: Refactor af_xdp_tx_zc Ariel Otilibili
2025-02-04 13:39     ` Maryam Tahhan
2025-02-06  1:09     ` Maryam Tahhan
2025-02-06  1:17       ` Maryam Tahhan
2025-02-06 18:06       ` Ariel Otilibili [this message]
2025-02-06  2:08     ` Maryam Tahhan
2025-02-06 17:56       ` Ariel Otilibili
2025-02-06 20:46 ` [PATCH v8 0/2] Fix use after free, and refactor af_xdp_tx_zc Ariel Otilibili
2025-02-06 20:46   ` [PATCH v8 1/2] net/af_xdp: Fix use after free in af_xdp_tx_zc Ariel Otilibili
2025-02-06 20:46   ` [PATCH v8 2/2] net/af_xdp: Refactor af_xdp_tx_zc Ariel Otilibili
2025-02-06 21:42     ` Stephen Hemminger

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to='CAF1zDgZ18Ov5ADzcFLgZoQB5RMVd8w=8-KpHQV6bGSe_QX44bg@mail.gmail.com' \
    --to=ariel.otilibili@6wind.com \
    --cc=ciara.loftus@intel.com \
    --cc=david.marchand@redhat.com \
    --cc=dev@dpdk.org \
    --cc=mtahhan@redhat.com \
    --cc=stable@dpdk.org \
    --cc=stephen@networkplumber.org \
    --cc=thomas@monjalon.net \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).